Statistical Inference for Gene-Level Analysis Based on Functional Linear Models

• Activity Number: 137 - Statistical Methods for Analyzing Genetic Variants and QTLs
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Olga Vsevolozhskaya and Adam Dugan* and David Fardo and Dmitri Zaykin
• Companies: University of Kentucky and University of Kentucky and University of Kentucky and National Institute of Environmental Health Sciences
• Keywords: gene-level test; functional data analysis; chi-squares mixture; reverse regression

Abstract

Testing genetic association of SNP sets (e.g., based on genes, haplotype blocks or annotation) has become a popular alternative to standard GWAS. In the reverse regression setting where predictors are regressed on outcomes, the SNP-set can be treated as a multivariate outcome, and a test for the genetic effect on multiple phenotypic traits may be accomplished. Standard multivariate techniques (e.g., Lawley-Hotelling trace or Wilks's Lambda) perform poorly in this setting due to not accounting for genetic correlation structure / linkage disequilibrium (LD). By contrast, functional data analysis techniques explicitly incorporate LD structure. Existing procedures for testing functional covariate effects include an F-type test proposed by Shen and Faraway (2004). Its null distribution is based on a weighted mixture of central chi-squares that can be approximated by the chi-squared distribution via moment matching. Here, we propose an alternative asymptotic approximation with improved robustness and apply it to test for an association between CSF analytes and Alzheimer's disease.